The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill relies on a 'kit' directory located outside its own folder structure. Instructions in SKILL.md and logic in engine/make.sh execute bash "$KIT/setup.sh" --voice to download and install approximately 1 GB of voice models.
[COMMAND_EXECUTION]: The skill workflow is orchestrated through shell scripts and Python tools that invoke multiple subprocesses, including ffmpeg for media processing and node for rendering. The rendering tool engine/tools/render.js utilizes Playwright to launch Chromium with the --allow-file-access-from-files flag, which is used to load local web assets for frame synthesis.
[CREDENTIALS_UNSAFE]: The text-to-speech utility engine/tools/tts_elevenlabs.py is designed to read the ELEVENLABS_API_KEY from the environment or a local .env file. While this is a standard practice for managing secrets in developer tools, it indicates that the skill processes sensitive authentication credentials.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided topics or questions to generate animated scripts. This creates a surface for indirect prompt injection, as the generated text influences the behavior of the audio and animation pipeline. Ingestion points: User input topics defined in SKILL.md. Boundary markers: Not explicitly present in the provided scripts. Capability inventory: Subprocess execution (ffmpeg, node) and network requests to ElevenLabs. Sanitization: No explicit sanitization or validation of the generated script content is observed in the provided files.

stick-figure